Figure 3
From: Protein embeddings predict binding residues in disordered regions
CAID2 binding benchmark. The Critical Assessment of protein Intrinsic Disorder round 2 (CAID215) assessed methods to predict binding regions in IDPRs (intrinsically disordered protein regions). Here, we showed IDBindT5 (method introduced in this paper) along with the SOTA methods ANCHOR214 and DeepDISOBind34, and a Random baseline (randomly assigning frequency of positives and negatives). Data set: CAID2 binding set with 78 proteins. The performance is captured by the four per-residue measures (scaled to [0,1]): precision (Eq. 1), recall (Eq. 2), balanced accuracy (Eq. 5), F1-Score (Eq. 6), and MCC (Eq. 7). Error bars reflect the 95% confidence interval (CI) of 1.96*standard errors. While IDBindT5 and DeepDISOBind reach higher numerical performance than ANCHOR2, not all differences are significant at 95% CI.
